Submitted by KAI LIU 3 JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation JavisVerse 10 1